We obtained data for the 369 species from three main sources: Birdlife international (BirdLife International, 2021), the website Birds of the World (https://birdsoftheworld.org) and Pigeons and Doves: A Guide to the Pigeons and Doves of the World (Gibbs et al., 2001). We extracted information on biological, ecological and environmental traits that are likely to influence extinction risk (see Buechley et al., 2019). We classified explanatory variables into four categories: “environment”, “behavior and life history”, “anthropogenic pressure” and “socio-political indicators” (Table 1). In order to estimate each species’ extinction risk, we relied on IUCN Red List categories and global population trends (BirdLife International, 2021). We classified species as threatened (critically endangered CR, endangered EN and vulnerable VU), non-threatened (near-threatened NT and least-concern LC), extinct (extinct EX and extinct in the wild EW), or data deficient. Because non-threatened species do not necessarily show stable or increasing population trends (Figure 1), we used global population trends of species (Rodrigues et al., 2006) as a second proxy of extinction risk. Species were thus classified as increasing, stable or decreasing according to BirdLife International (2021).

In order to quantify scientific attention given to each columbid species, we relied on the number of peer-reviewed articles retrieved from an extensive bibliographic search on all databases of the Web of Science (conducted by mid-2021). We performed searches using current and past binomial scientific names, vernacular names in English, French and Spanish and all recognized synonyms. To compile all possible names of species, we used the websites Avibase (https://avibase.bsc-eoc.org) and Birds of the World. We considered several “research areas”, as defined in the Web of Science, as all studies bringing ecological knowledge about a species may be of help for its conservation (Courchamp et al., 2015). Each search was conducted using the following procedure:

“TS=”

(in order to search following terms in title, abstract and keywords of articles)

+

Full binomial scientific names, vernacular and all recognized synonyms in independent quotes, separated by “OR”

+

“AND SU=(“Ecology” OR “Biodiversity and Conservation” OR “Ornithology” OR “Evolutionary Biology” OR “Zoology”)”

(to limit searches to studies related to following science disciplines: Ecology, Conservation Biology, Ornithology, Evolutionary Biology/Ecology and Zoology)

+

“NOT SU=(“Veterinary Sciences” OR “Neurosciences and Neurology” OR “Anthropology”)”

(to exclude any study related to veterinary sciences, neurosciences and neurology, and anthropology).

We also performed a global search in order to estimate national contributions to research on Columbidae, assessed from the affiliations of authors of peer-reviewed articles. To that end, we used the same search nomenclature, except that we replaced names of species and their synonyms by “Columbidae”, “Columbiformes”, “Pigeon” and “Dove”, separated by the Booleans operator “OR”. In order to exclude all articles on Pigeon pea (Cajanus cajan), we added: NOT TS=(“Pigeon*pea*” OR “Cajanus cajan”).

We used QGIS (3.10.11) to calculate spatial statistics for explanatory variables used in our analyses (Table 1 and Supplementary Material 1, Method A1). We obtained various raster and polygon files for historical species range from several sources (Table 1), and rasterized polygons representing area distribution of species using scientific attention (log[number of articles + 1]) as value for the raster unit (square of 400km²). Since, polygons of several species may overlap, we beforehand calculated the mean of scientific attention per raster unit.

The RF model classifying species either as non-threatened or threatened correctly classified 87% of the species based on the 38 explanatory variables used (Table 1). RF model sensitivity and specificity respectively indicated that 87% of actual non-threatened species and 82% of actual threatened species were correctly identified. Cohen’s kappa statistics and TSS further indicated that the classification of species significantly differed from a random classification [Cohen’s kappa = 0.51, P< 0.0001; TSS = 0.70], while the OOB error was 13%. The eight most important variables influencing extinction risk were in decreasing order: species historical range size, impact of invasive alien mammals, distance to closest mainland, altitude distribution of the species, HFP, human density, proportion of species historical range in current biodiversity hotspots, and species body mass. Based on the 95% CI comparison, only historical geographic range size had a relative importance significantly higher than the other variables (Figure 2A, see Supplementary Material 2, Figure A4 for all variables importance values). Based on partial dependence plots, species with the smallest historical range size, exposed to invasive alien mammals, living the farthest to mainland and/or at the lowest altitudes were associated to a greater risk of extinction (Figure 3A). Although other variables were relatively important in the prediction of species’ risk extinction based on the RF model, their variation did not seem to be directly associated to extinction risk (Figure 3A; see Supplementary Material 2, Figure A5 for all variables).

The RF model correctly classified 73% of species according to population trends, with a OOB error of 27%. In addition, the percentage of true positive was of 74%, 69% and 64% for decreasing, stable and increasing species, respectively. On the other hand, the percentage of true negative was of 80%, 89% and 94% for decreasing, stable and increasing species’ population trends, respectively. Although goodness-of-fit was lower than for the previous model, Cohen’s kappa still indicated a species classification differing from a random classification [Cohen’s kappa = 0.48, P< 0.0001]. The eight most important variables predicting population trends were in decreasing order: human density within the species historical geographic range distribution, species body mass, HFP, species historical geographic range size, species habitat breadth, GSI, species altitude distribution, and HDI (Figure 2B). Based on the comparison of 95%CI, these eight variables did not seem to significantly differ in terms of importance (Figure 2B, see Supplementary Material 2, Figure A4 for all variables importance values). Based on partial dependence plots, risk of population decline was significantly lower for species with a low body weight living in areas characterized by high human density (Figure 3B). In addition, a higher risk of decline was associated with species with narrow habitat breadth and living at low altitudes (Figure 3B). Although other variables were important in predicting species’ risk of decline based on the RF model, partial dependent plots suggest that they are not directly associated to it (Figure 3B; see Supplementary Material 2, Figure A6 for all variables).

Overall, our results tend to agree with previous findings (Walker, 2007) about the main drivers of extinction risk. However, we show that the use of both IUCN status and population trends are complementary in the identification of these drivers. Using both of them allowed us to identify several biological and ecological traits that make columbid species more prone to extinction. First, range-restricted species appeared more prone to extinction than more geographically widespread ones. Such relationship might seem obvious when using IUCN status as a proxy of extinction risk, because geographic range size is considered in its assessment (Fisher and Owens, 2004). However, geographic range size of columbid species was also positively associated with risk of population decline, thus confirming this factor as an important predictor of global extinction risk. Although the impact of geographic range size on extinction risk is complex, small range size is often associated with restricted ecological niche and small population size. Indeed, the number of habitats or resources required for survival and reproduction of individuals and, hence, population growth, often covary with geographic range size (Gaston, 2003; Peterson et al., 2011; Saupe et al., 2015). Moreover, large-range species are overall less likely to be affected by human disturbance or destruction of their natural environment, or by stochastic events, in their entire geographic range at once (Gaston, 2003; Fisher and Owens, 2004). However, some particular migrating columbid species, despite having a large range size, might be vulnerable to disturbance or damage occurring in their breeding, wintering or stop-over areas, as shown in the case of the Woodpigeon (C. palumba; Schumm et al., 2022; but see Bendjoudi et al., 2015) or the Turtle Dove (S. turtur; Eraud et al., 2013; Schumm et al., 2021). The negative relationship between extinction risk and geographic range size has been evidenced in a large range of terrestrial and marine vertebrate species (Davidson et al., 2012; Di Marco et al., 2014; Böhm et al., 2016; Buechley et al., 2019), and at different geological time scales (Harnik et al., 2012).

Second, island columbid species appeared to be more prone to extinction than continental ones, and even more so the further the distance of islands to mainland. Indeed, in addition to usually having smaller geographic range size than continental species, island-species have evolved in isolation, under particular selection pressures (e.g., relaxed predation and parasitism), making them particularly vulnerable to disturbances such as overexploitation and any kind of negative interactions with invasive species (Purvis et al., 2000a; Purvis et al., 2000b; Boyer, 2010; Loehle and Eschenbach, 2012). Introduced pathogens may also have a significant negative impact on the persistence of isolated populations. For instance, a pathogenic strain of Trichomonas gallinae, which might have been introduced in Mauritius Island, increased adult and nestling mortality in the endangered Pink Pigeon, Neosoenas mayeri (Swinnerton et al., 2005; Bunbury et al., 2007; Bunbury et al., 2008). In mammals and birds, 95% of known extinctions concerned island-species (Loehle and Eschenbach, 2012). However, smaller geographic range size of most insular species might be the most important variable in explaining species extinction risk (Manne et al., 1999). Among columbid species, all known extinct species inhabited islands, excepted for the Passenger Pigeon (Ectopistes migratorius), suggesting that being an island-species is at least as much important as geographic range size, especially when adding other sources of extinction such as overexploitation, invasive species or climate change (Courchamp et al., 2003; Courchamp et al., 2014; Wood et al., 2017). For instance, the two largest known and now extinct columbid species, the iconic Dodo, Raphus cucullatus, and the Rodrigues Solitaire, Pezophaps solitaria, were both endemic to Mauritius islands and flightless. Their extinction was probably due to a combination of factors, including large body size, island endemism, low population growth, small population size and small geographic range size, making them particularly vulnerable to habitat destruction, hunting and invasive species (Duncan et al., 2002; Fisher and Owens, 2004; Hume, 2006; Hilbers et al., 2016). Our results indeed suggest that the extinction risk of columbid species increases with their body mass, as shown in diverse taxa (Gaston and Blackburn, 1995; Davidson et al., 2009; Fritz et al., 2009; Böhm et al., 2016). The influence of body size could actually be multifactorial, and be exacerbated by other external factors like overexploitation. Indeed, larger species are often preferred as game species (Duncan et al., 2002; Whytock et al., 2016; Ripple et al., 2017).

Although hunting has been recognized among the most important threats within Columbidae, concerning 77% of threatened species (Walker, 2007), and that inferential statistics performed in this study (see Supplementary Material 2, Table A1–A2) showed that threatened species are significantly more exposed to it, our models did not identify hunting-related drivers (i.e., subsistence use and sport hunting) as important variables explaining extinction risk. This does not mean that the impact of hunting exposure is null, but that other variables provide a better discriminant classification into RF models to evaluate extinction risk of columbids. It is obvious that hunting has a direct influence on the mortality rate of harvested populations, and, hence, can be a contributing driver to extinction risk for some species (e.g., the European Turtle Dove, Streptopelia turtur, Lormée et al., 2020), especially when it is not regulated or not performed in a sustainable way. However, the assessment of extinction risk requires a more comprehensive approach, considering various drivers and threats, as discussed in management recommendations for the European Turtle Dove (Bacon et al., 2023). The global influence of hunting might thus be lower than that of other threats, such as habitat loss and invasive species. However, its impact on game populations might be exacerbated when combined with such threats (Romero-Muñoz et al., 2019; Romero-Muñoz et al., 2020; Bogoni et al., 2022). For instance, the extinction of the iconic Passenger Pigeon was caused by both unregulated commercial hunting and significant destruction of its breeding habitat (i.e., a decrease estimated about 53%; Stanton, 2014). Furthermore, a more precise metrics of hunting pressure would probably be more informative than mere exposure to hunting (Bellard et al., 2022) and could have enhanced the performance of our RF models. Unfortunately, such information is unavailable for a large majority of game columbids.

Habitat breadth also appeared to influence extinction risk, confirming a general trend (Norris and Harper, 2004; Devictor et al., 2008; Clavel et al., 2011; Callaghan et al., 2019). Although ecological specialization does not always limit the capacity of species to persist (Benito et al., 2020), it might be evolutionary disadvantageous in environments exposed to changes and perturbations (Davies et al., 2004; Colles et al., 2009; Clavel et al., 2011; Ducatez et al., 2014; Morelli et al., 2020). Indeed, specialist species often tend to be associated with small population size, restricted geographic range size and limited use of resources and habitats, what may make them particularly vulnerable when exposed to environmental changes and disturbances (McKinney, 1997; Davies et al., 2004; Colles et al., 2009).

Not surprisingly, our analysis points to alien invasive species as an important factor affecting the risk of extinction of columbid species (Walker, 2007). Exotic species are indeed a major threat to biodiversity, as a result of predation, parasitism, competition, hybridization with native species and the disturbance of native ecosystems (Mack et al., 2000; Sax and Gaines, 2008; Walsh et al., 2012). Invasive mammals, especially rodents (e.g., rats Rattus sp.) and cats (Felis catus), are the most damaging group of alien species across all ecosystems. Both are responsible of more than 30% of all known contemporary extinctions, especially on islands (Doherty et al., 2016). Invasive species can predate native columbid species (Rodriguez-Estrella et al., 1991; Tidemann et al., 1994; Zarzoso-Lacoste et al., 2016; Lamelas-López et al., 2020), compete with them for resources (James and Clout, 1996; Bond et al., 2019), or transmit pathogens (Stimmelmayr et al., 2012). Based on IUCN data, 9% of all columbid species have been reported to be exposed to invasive rats, 8% to cats, and 12% to other mammals (IUCN, 2021). Among non-threatened species, 3% of species have been reported to be exposed to invasive mammals, compared to 35% for threatened species and 77% for extinct species. However, the precise impact of invasive mammal species on the demography of columbid species of patrimonial interest in the wild remains to be quantitatively assessed in order to assess to what extent it compares in importance with hunting pressure or habitat loss and thus provide a more comprehensive and integrative approach to the management of species/populations.

Although global warming is supposed to increase extinction risk among species living at high altitude (Williams et al., 2003; Shoo et al., 2005; Sekercioglu et al., 2008; Bässler et al., 2010; La Sorte and Jetz, 2010), we found no evidence for a positive relationship between extinction risk and elevation in columbids. The consequences of global warming on extinction risk at higher elevation might not be felt yet (Bellard et al., 2012) in columbid species, or might have been hidden in our analysis by variation in habitat breadth and more intense human activities at lower elevations. Indeed, landscape heterogeneity tends to decrease with increased elevation, and species found at elevation tend to have a higher degree of ecological specialization than those found at lower elevation (Reif et al., 2013; Rivas-Salvador et al., 2019). Lower elevations are also associated with higher anthropogenic transformation, inducing various negative effects on local biodiversity (Ellis et al., 2010; Reif et al., 2013; White and Bennett, 2015; Smith et al., 2019). Further attention should then be given to the influence of elevation on the demographic trends of columbid species in relation to climate change and habitat modification.

Several other drivers of anthropogenic origin appeared to contribute, directly or indirectly, to increase the extinction risk of columbid species, such as low levels of governmental stability and poor socioeconomics and/or the loss and disturbance of natural habitats through the development of HFP. HFP provides an index of human conversion of natural terrestrial ecosystems based on the measure of eight different anthropogenic pressures related to urbanization, agriculture and human density (Venter et al., 2016), and can therefore be regarded as a proxy of habitat degradation or loss. Yearly variations in HFP have been found to be a good predictor of extinction risk in terrestrial mammals (Di Marco et al., 2018). Although some columbid species flourish in landscapes modified by humans, such as the widespread Feral Pigeon, Columba livia domestica (Carlen et al., 2021), the Zenaida Dove, Zenaida aurita (Wolff et al., 2018), or the invasive Eurasian Collared-Dove, Streptopelia decaocto (Luna et al., 2018), most species rely on their natural habitats to persist. For instance, the occupancy of the Kereru, Hemiphaga novaeseelandiae, an endemic New Zealand Pigeon, has been affected by habitat loss resulting from European settlement in New Zealand (Carpenter et al., 2021). Following Walker (2007), the present study confirms that the loss of natural habitat is the greatest cause of extinction and populations decline in columbid species.

Governmental and socioeconomic factors were also found to affect the extinction risk of columbid species, with species inhabiting in developing countries being more exposed to extinction risk. Public policies can have negative or positive influences on biodiversity, through supporting or not reliable environmental and conservation projects and by protecting or not most vulnerable species and/or sensitive ecosystems (Chapron, 2014; Ripple et al., 2016). For instance, the Pink Pigeon, Nesoenas mayeri, have been downlisted twice, in the IUCN red list, from critically-endangered to vulnerable thanks to conservation projects lead by conservationists, private foundations and the Mauritian government (Tatayah, 2019). Globally, the socioeconomic level of countries tends to influence positively the persistence of species (Forester and Machlist, 1996; Dietz and Adger, 2003). For instance, Bösch (2021) showed that the level and speed of countries’ economic-institutional development tend to be negatively associated with illegal logging, and hence, with habitat loss and biodiversity decline. The same is true for wildlife trafficking and hunting pressure, especially in developing countries where wildlife can be the sole source of proteins for isolated and/or low-income communities (Robinson and Bennett, 2002; Fa et al., 2003; Fa et al., 2009; Borgerson et al., 2016; Lescuyer and Nasi, 2016). Legal and illegal hunting can be practiced to obtain food or increased social status, or as a recreational or traditional activity (Phelps et al., 2016; Stirnemann et al., 2018). Indeed, wildlife meat consumption is not always associated with poverty and can also be a sign of wealth (Fa et al., 2009). This is for instance the case in the Pacific islands where poor people hunt both the Pacific Pigeon, Ducula pacifica, and the Tooth-billed Pigeon, Didunculus strigiristris, to sell their meat to high-income populations (Phelps et al., 2016; Stirnemann et al., 2018).